Nearly 746,000 people sustain a spinal cord injury (SCI) every year leading to impairment or even complete loss of motor functions. Motor Brain-Machine Interfaces (BMIs) translate brain neural signals into commands to external effectors directly providing patients with control over orthoses, prostheses, or over their own limbs using electrical stimulation.

The project will address the challenge task of developing a real-time decoding model that will allow controlling the movement of the arms of the exoskeleton developed at Clinatec.

To reach this goal, the two main objectives of the post-doctorate are:

• BMIs are ‘human-in-loop’ systems. Recorded brain neuronal signal includes neuronal feedback. It makes offline training of decoding model less efficient compared to closed loop decoder adaptation. Real-time model update requires incremental/continual decoder learning. To date, the exoskeleton works with an incremental machine learning algorithm but recent results have shown the interest of deep learning at the output of the decoder. However, deep learning algorithms suffer from catastrophic forgetting. Indeed, when an artificial neural network learns new information, its synaptic weights adjust to this information, without forgetting the previously informed information. The development of incremental ANN learning for Adaptive-BMI (Ada BMI) framework will be the first breakthrough of the project.

• The second objective concerns the decoder performances, more particularly the extraction of more informative features. Indeed, these features evolve over time with for instance cross frequency coupling. Taking into account this kind of phenomenon will allow us to input better features in the incremental learning algorithm.

Both improvements will lead to a decrease of the training time sessions and so an improved comfort for patients. Thus, the subject is a the interface of three labs of the CEA: CEA-LIST, Clinatec and Neurospin.

The candidate should have completed a PhD in Computer Science, Machine Learning.

Knowledges and experiences in some or all of the following fields will be an asset during the position:

• Machine Learning / Deep learning
• Applied mathematics (probability / statistics)

Good programming practice in Python (Tensorflow, with some basic GPU environment knowledges). Applicants should master written and spoken English.

A brief description of the PhD thesis, a publication list and some recommendations should be included to your application